The Dark Side of the Force: When computer simulations lead us astray and "model think" narrows our imagination. (revised version, October 2006)

Eckhart Arnold

1 Introduction
2 Different aims of computer simulations in science
3 Criteria for explanatory simulations
4 Examples of Failure: Axelrod style simulations of the “evolution of cooperation”
5 Conclusion

1 Introduction

Computer simulations have become a popular tool in various branches of science, including even the social sciences. The reasons are easy to understand: Computer simulations provide a simple and yet powerful tool to explore the implications of theoretical assumptions. They are cheaper than experiments and often easier to construct and to handle than mathematical models. At the same time they confine the realm of what can be modeled only to what can be described algorithmically, which gives them a very broad scope. With this tool at hand it should be possible to bring into the reach of exact treatment even such questions that have traditionally seemed to defy the use of formal methods.

However, upon closer inspection it becomes apparent that computer simulations do not always deliver what they promise. Often they remain in the state of purely theoretical “toy simulations” and never get to the ground of empirical testability. In the following, I will first try to put forward a few straightforward criteria for proper explanatory computer simulations. After that I will analyze some examples of computer simulations that fail to meet these criteria and I will try to point out the consequences this failure has.

Much of what will be discussed in the following concerns the limits of formal modeling in general and not just computer simulations. Yet it seems that the question is more urgent in the case of computer simulations. For, in the case of the relatively new technique of computer simulations the sensitivity of the scientific community for the need of empirical justification or, more general, the feel for what are good practises and what are bad practises when employing this new tool has not yet grown quite mature. I have a faint hope that the following discussion of the limits of computer simulations will help to develop this sort of sensitivity, even if in some places it may turn out to be wrong or overly critical.

t g+ f @